Walks you through building agent skills with the right structure: lean main instructions, supporting files in references and assets, and optional skillgrade evals for regression testing. The focus is progressive disclosure, so the agent only reads what it needs when it needs it. Use this when you're spinning up a new skill directory or tightening up metadata so the router can find it. The inline shell command advice is smart: only snapshot state when it actually improves the prompt, and keep it deterministic. If you're maintaining a skill library or building tools for Claude, this codifies the workflow so you're not reinventing structure every time.
npx skills add https://github.com/cap-go/capgo-skills --skill skill-creator